Earthquake damage assessment in Nepal using machine learning and SHAP: a regional analysis of building vulnerabilities

Earthquake disaster is burning issue of the world including Nepal. It has became crucial phenomenon of global society. In this context his research paper examines the sociological aspects of earthquake disaster-resilient housing construction initiatives in landless communities affected by the 2015 Gorkha earthquake in Nepal. It highlights the significant outputs and outcomes of local housing strategies that leverage indigenous materials, knowledge, and social support. Utilizing a descriptive research methodology, data was gathered through close-ended and open-ended questionnaires, case studies, and field observations across various impacted communities. Secondary data sources included existing literature on landless housing reconstruction and community-based initiatives. The findings reveal that the landless communities suffered greatly due to seismically vulnerable structures, lack of preparedness, and insufficient use of critical seismic features. These communities often lack access to vital information that could mitigate disaster risks. The study further explores how residents have innovatively rebuilt their homes by integrating safety measures, utilizing salvaged materials, and combining traditional knowledge with scientifically validated techniques. This approach not only addresses immediate housing needs but also fosters long-term resilience against future disasters. Keywords: adaptive, climate-change, earthquake disaster-management, resilient, social-support

Information on Years of Life Lost (YLL) is instrumental in determining the mortality impact of disasters at the population level. YLL due to the 2015 Gorkha earthquake in Nepal, along with associated demographic disparities, has not yet been systematically studied. We therefore estimated YLL due to the 2015 Gorkha earthquake in Nepal and examined disparities by age, sex, and district. Earthquake-specific deaths were sourced from the 2017 Nepal Police Report and disaggregated from broad age groups into 5-year intervals using a smooth, shape-preserving cubic interpolation approach that combines district-level population structures with age-specific mortality patterns. YLL were calculated by multiplying deaths in each age group by the remaining life expectancy from the 2021 Global Burden of Disease (GBD) reference life table. A sensitivity analysis was performed using the 2019 national life table for Nepal. Reporting follows the Standardised Reporting Of Burden Of Disease (STROBOD) statement. The earthquake caused an estimated 8,951 deaths across 41 districts, resulting in 473,280 YLL and 2,685 per 100,000. YLL rates were higher in females (2,998 per 100,000) than in males (2,365 per 100,000). Individuals under 20 years accounted for 55% of total YLL. Male-to-female YLL ratios ranged from 0.15 (Parsa) to 6.14 (Solukhumbu). YLL rates were 27% lower for males and 21% lower for females when based on the Nepal rather than the GBD life table. YLL due to the 2015 Gorkha earthquake varied considerably by age and sex, and estimates were influenced by the choice of life table. The 2015 Gorkha earthquake ranked among the top five leading causes of YLL in Nepal in 2015. Future research should integrate morbidity data, refine age-specific mortality estimates, and strengthen disaster risk reduction strategies targeting vulnerable groups.

Abstract. This study introduces a new approach to multi-hazard risk assessment, leveraging hypergraph theory to model the interconnected risks posed by cascading natural hazards. Traditional single-hazard risk models fail to account for the complex interrelationships and compounding effects of multiple simultaneous or sequential hazards. By conceptualising risks within a hypergraph framework, our model overcomes these limitations, enabling efficient simulation of multi-hazard interactions and their impacts on infrastructure. We apply this model to the 2015 Mw 7.8 Gorkha earthquake in Nepal as a case study, demonstrating its ability to simulate the primary and secondary effects of the earthquake on buildings and roads across the whole earthquake-affected area. The model predicts the overall pattern of earthquake-induced building damage and landslide impacts, albeit with a tendency towards over-prediction. Our findings underscore the potential of the hypergraph approach for multi-hazard risk assessment, offering advances in rapid computation and scenario exploration for cascading geo-hazards. This approach could provide valuable insights for disaster risk reduction and humanitarian contingency planning, where the anticipation of large-scale trends is often more important than the prediction of detailed impacts.

Co-seismic response of water level in the Jingle well (China) associated with the Gorkha Nepal (Mw 7.8) earthquake

Abstract. Coseismic avalanches and rockfalls, as well as their simultaneous air blast and muddy flow, which were induced by the 2015 Gorkha earthquake in Nepal, destroyed the village of Langtang. In order to reveal volume and structure of the deposit covering the village, as well as sequence of the multiple events, we conducted an intensive in situ observation in October 2015. Multitemporal digital elevation models created from photographs taken by helicopter and unmanned aerial vehicles reveal that the deposit volumes of the primary and succeeding events were 6.81 ± 1.54 × 106 and 0.84 ± 0.92 × 106 m3, respectively. Visual investigations of the deposit and witness statements of villagers suggest that the primary event was an avalanche composed mostly of snow, while the collapsed glacier ice could not be dominant source for the total mass. Succeeding events were multiple rockfalls which may have been triggered by aftershocks. From the initial deposit volume and the area of the upper catchment, we estimate an average snow depth of 1.82 ± 0.46 m in the source area. This is consistent with anomalously large snow depths (1.28–1.52 m) observed at a neighboring glacier (4800–5100 m a.s.l.), which accumulated over the course of four major snowfall events between October 2014 and the earthquake on 25 April 2015. Considering long-term observational data, probability density functions, and elevation gradients of precipitation, we conclude that this anomalous winter snow was an extreme event with a return interval of at least 100 years. The anomalous winter snowfall may have amplified the disastrous effects induced by the 2015 Gorkha earthquake in Nepal.

The Himalaya has experienced large damaging earthquakes over the past few centuries, most recently the damaging 25 April 2015 M 7.8 Gorkha earthquake in Nepal. Because of the continued earthquake risk presented by the continental collisional plate boundary at the Main Himalayan thrust and the high population densities in the region, collecting and processing data related to recent large earthquakes in this region is critically important for improving our understanding of the regional tectonics and earthquake hazard. Following the 2015 Gorkha earthquake, we deployed a National Science Foundation-funded rapid-response aftershock network known as the Nepal Array Measuring Aftershock Seismicity Trailing Earthquake network across the rupture area for 11 months beginning 7 weeks after the mainshock. The network consisted of 41 broadband and short-period seismometers, and 14 strong-motion sensors at 46 sites across eastern and central Nepal. The network spanned a region approximately 210 km along strike by 110 km across strike with a station spacing of 20–25 km. In this article, we report lessons learned from this deployment as well as details of the publicly accessible dataset including data recovery, data quality, and potential for future research.

Background Nepal has a history of regular natural hazards, including significant earthquakes and floods. Previous studies have shown that females, particularly younger females, are more likely to die due to disasters compared to males. This study aimed to explore gender differences in mortality - measured in Years of Life Lost (YLL) - caused by the 2015 Gorkha earthquake across 41 districts of Nepal. Methods Mortality counts were derived from the Nepal Police Report. YLL was estimated by multiplying mortality counts (stratified by age, gender, and district) by age-conditional life expectancy from the 2019 Global Burden of Disease reference life table. Both absolute numbers and rates of YLL per 100,000 population were calculated. YLL per deaths, stratified by age-gender-and-location, were estimated and compared. Male-to-female YLL rate ratios were estimated as well, and compared across districts. Results The 2015 Gorkha earthquake resulted in 8,950 deaths (females: 56.1% versus males: 43.9%), accounting for 526,617 YLL. Males had a slightly higher number of YLL per death (59.2) compared to females (58.5); this is mainly due to the different number of deaths between males and females. YLL per deaths ranged from 65.4 (Rautahat district) to 0 (in several districts) among males and, from 65.3 (Rautahat district) to 0 (in several districts) among females. The YLL rate per 100,000 population was higher for females (3,158 per 100,000) than for males (2,575 per 100,000). Approximately 40% of the districts had male-to-female YLL rate ratios of less than one, which can be partly explained by the gender distribution of the population at rik of dying due to earthquake. Conclusions The study found relatively higher YLL per death for males and higher YLL rates for females. This underscores the necessity to further investigate the factors that contribute to fatalities from earthquakes, especially in lower middle income countries such as Nepal. Key messages • Gender disparities in mortality following the 2015 Gorkha earthquake in Nepal were addressed using Years of Life Lost (YLLs); males had higher YLL per death, while females had higher YLL rates. • Exploring the factors that contribute to earthquake-related fatalities should be a priority for disaster mitigation strategies.

The April 25, 2015 Mw 7.9 Gorkha earthquake in Nepal was characterized by a peak slip of several meters and persisting aftershocks. We report here that, in addition, a dense seismic swarm initiated abruptly in August 2017 at the western edge of the afterslip region, below the high Himalchuli-Manaslu range culminating at 8156 m, a region seismically inactive during the past 35 years. Over 6500 events were recorded by the Nepal National Seismological Network with local magnitude ranging between 1.8 and 3.7 until November 2017. This swarm was reactivated between April and July 2018, with about 10 times less events than in 2017, and in 2019 with only sporadic events. The relocation of swarm earthquakes using proximal temporary stations ascertains a shallow depth of hypocenters between the surface and 20 km depth in the High Himalayan Crystalline slab. This swarm reveals an intriguing localized interplay between orogenic collapse and stress adjustments, involving possibly CO2-rich fluid migration, more likely post-seismic slip and seasonal enhancements.

In this article, we created a well-resolved aftershock catalog for the 2015 Gorkha earthquake in Nepal by processing 11 months of continuous data using an automatic onset and hypocenter determination procedure. Aftershocks were detected by the NAMASTE temporary seismic network that is densely distributed covering the rupture area and became fully operational about 50 days after the mainshock. The catalog was refined using a joint hypocenter determination technique and an optimal 1D velocity model with station correction factors determined simultaneously. We found around 15,000 aftershocks with the magnitude of completeness of ML 2. Our catalog shows that there are two large aftershock clusters along the north side of the Gorkha–Pokhara anticlinorium and smaller shallow aftershock clusters in the south. The patterns of aftershock distribution in the northern and southern clusters reflect the complex geometry of the Main Himalayan thrust. The aftershocks are located both on the slip surface and through the entire hanging wall. The 1D velocity structure obtained from this study is almost constant at a P-wave velocity (VP) of 6.0 km/s for a depth of 0–20 km, similar to VP of the shallow continental crust.

Housing is vital in facilitating a return to normality following a disaster; however, it remains one of the most challenging and problematic areas of post-disaster assistance. There is a pressing need to unpack “Build Back Better” aspirations to understand the drivers of safe housing reconstruction. The objective of this research was to understand the influence of technical assistance and sources of funding on household perceptions of housing safety. Binomial logistic regression was used to analyse survey data collected from 711 households three and half years after the 2015 Gorkha Earthquake in Nepal in two affected districts. We found that only 55% and 60% of households in the selected districts of Gorkha and Okhaldhunga, respectively, felt their home was safe in case of a large future earthquake. The use of demonstration houses in communities resulted in higher odds of safe perceptions, while door-to-door technical assistance was associated with lower perceptions of safety in Gorkha. In contrast, in Okhaldhunga, household reconstruction orientations and short training resulted in lower odds of safe perceptions. The funding source for housing reconstruction did not correlate with perceptions of safety in Gorkha, while government funding and household savings had positive correlations in Okhaldunga. Our findings suggest a possible link between reconstruction financing, technical assistance, and the perceptions household form out of recovery. Organisations seeking to support housing reconstruction can potentially leverage tailored technical assistance and funding sources as points of entry to influence household demand for safer construction.

The ability of a community to withstand and recover from adversities including natural and man-made disasters has emerged as a major policy issue in recent years. This research aims to assess the role of institutional initiatives in building resilient communities and their response to natural disasters like the Gorkha earthquake in Nepal in 2015. The work is based on data collected from primary and secondary sources along with field observations. It is evident that resilient communities are equipped with greater coping capacities in the face of natural disasters and have reduced vulnerability to future hazards. Institutional capacity building and resilient construction including the School Earthquake Safety Program ensured better disaster preparedness. The traditional open spaces and building designs added to the structural resilience. There is, however, a need to build back better and to communicate earthquake-resistant designs to the affected communities.

The April 25, 2015 Gorkha Earthquake of Magnitude 7.8 in Nepal damaged about seven hundred thousand buildings. The main typologies of buildings in the affected area are stone masonry with mud mortar, some buildings with stone and brick masonry with cement/sand mortar and few reinforced concrete buildings with masonry infill. Among the damaged buildings, about 96% of the buildings were masonry and about 4% reinforced concrete buildings with masonry infill. This study conducted detailed damage assessment of over 150,000 buildings of different type of masonry and reinforced concrete (rc) buildings in Nepal. First, the buildings were classified according to different structural types like adobe, stone in mud, brick in mud, stone in cement, brick in cement, wood, bamboo, rc and others. Other important parameters like type of floors and roofs and occupancy of the buildings were noted before starting the detailed damage assessment of structural elements. Damage to overall building as well as to different structural/non-structural elements was categorized into four different categories mainly overall hazard, structural hazard, non-structural hazard and geotechnical hazard. The damage level to different structural/non-structural elements was assigned from insignificant damage to extreme damage in three categories considering the severity of damage like crack widths, delamination, tilting etc. In addition to the severity of damage, extent of damage to that particular element of different severity was also noted. Each type of damage with different severity was estimated in terms of extent like less than 1/3rd of the total area, 1/3rd–2/3rd and more than 2/3rd. Considering the damage severity and extent, overall damage grade to the building was assigned. Finally, based on the damage grade and extent of damage, recommendation for the building, either to demolish, repair and retrofit or just repair was recommended. This study further analyzes the main type of damage to different categories of the buildings and finds out critical factors to be considered for making them earthquake resistant. Existing traditional earthquake resistant elements like wooden bands and their effectiveness on earthquake safety of masonry buildings are further studied. It is found that, corner separation, diagonal cracking, out of plane failure, in-plane flexural failure and delamination are the main type of damage to masonry buildings while soft-story damage, joint failure, lap splice, columns shear failure, beam failure and infill walls failure are the main types of damages to non-engineered rc-buildings.

Building vulnerability assessment in seismic areas using ensemble learning: A Nepal case study

Landslide hazard mapping is critical for disaster risk reduction and resilient planning in vulnerable regions. Whilst the existing methods for Landslide hazard mapping (LHM) provide good predictability, they still lack in analysing the impact of parameters on the hazard risk. Although Explainable AI (XAI) application enhances this capability, its use in LHM analysis is still limited. This study presents an interpretable framework for evaluating landslide hazards in Raigad district, Maharashtra, that integrates machine learning models, Random Forest (RF) and Support Vector Machine (SVM) with statistical techniques, Frequency Ratio (FR) and Shannon Entropy (SE). Sixteen conditioning factors, selected through multicollinearity screening and feature‐selection methods and spatially validated inventory of 174 datasets of landslides and non‐landslides each, were used for analysis. Model performance was assessed using evaluation metrics. The RF model achieved the highest AUC‐ROC of 0.90, followed by SVM (0.81), SE (0.80), and FR (0.79). To identify critical parameters affecting landslide vulnerability, XAI methods such as SHapley Additive exPlanations (SHAP) and partial dependence plot analysis were applied. Slope angle emerged as the most dominant predictor of landslide risk. The generated hazard maps can offer more useful insights into the categorisation of land based on the degree of landslide risk. These maps can be used in development planning to help with infrastructure design, land‐use zoning, and prioritisation of high‐risk locations for mitigation.

This study examines diet as a key risk factor for sleep disorders and integrates physiological indicators to develop a machine learning (ML)-based model for targeted public health interventions. Data from 5158 2011 to 2014 National Health and Nutrition Examination Survey (NHANES) participants were analyzed. Dietary, lifestyle, and physiological variables used to build sleep disorder prediction models with random forest, extreme gradient boosting, light gradient boosting machine, and logistic regression. Model interpretability was assessed using Shapley additive explanations (SHAP). Key predictors were further analyzed using progressive modeling and least absolute shrinkage and selection operator (LASSO) regression. All ML models showed acceptable-to-excellent discrimination (area under the receiver operating characteristic curve: 0.744–1.000), with light gradient boosting machine achieving the highest performance (area under the receiver operating characteristic curve = 1.000). SHAP analysis showed that dietary inflammatory index (DII), body mass index (BMI), and age were positively associated with sleep disorder risk, while mean arterial pressure was negatively associated. In progressively adjusted logistic regression models, BMI was consistently positively associated with sleep disorders (model 3 odds ratio [OR] = 1.065, 95% confidence interval [CI]: 1.050–1.080; P < .001), whereas DII was associated with sleep disorders primarily in less-adjusted models (model 1 OR = 1.099, 95% CI: 1.035–1.168; P = .002; model 2 OR = 1.072, 95% CI: 1.004–1.145; P = .037). To further identify which dietary components driving the DII-related signal were most relevant to sleep disorder risk, we applied LASSO to the nutrient components of DII, which selected iron, carbohydrates, and total fat as the major contributors to the diet-related sleep disorder risk profile. An interpretable ML model based on National Health and Nutrition Examination Survey data demonstrated good discrimination for sleep disorders and consistently highlighted BMI and DII as central correlates. SHAP and LASSO further translated these associations into clinically interpretable dietary signals, including iron, carbohydrate, and total fat intake within the DII framework, supporting screening-oriented risk profiling and prioritization of individuals for further sleep evaluation and targeted nutrition assessment.